3
$\begingroup$

I have seen other posts in this forum but didn't find any convincing answer.

Random Forest has an another way of tuning hyperparameter via OOB by design. OOB and CV are not the same as OOB error is calculated based on a portion of trees in Forest rather by full Forest.

So what are the advantages and disadvantages of using OOB instead of a CV?

Is getting to train on more data by using OOB correct to say?

$\endgroup$
3
$\begingroup$

OOB samples are a very efficient way to obtain error estimates for random forests. From a computational perspective, OOB are definitely preferred over CV.

Also, it holds that if the number of bootstrap samples is large enough, CV and OOB samples will produce the same (or very similar) error estimates. Thus, if you perform many bootstrap samples, I would recommend performing cross-validation along the way with OOB samples until the OOB error converges.

$\endgroup$
  • $\begingroup$ I'm not sure to correctly understand this sentence: "Thus, if you perform many bootstrap samples, I would recommend performing cross-validation along the way with OOB samples until the OOB error converges." You mean that when we have a lot of training data, we should mix cross-validation and OOB? $\endgroup$ – Pierre Apr 18 '18 at 6:57
  • $\begingroup$ Stumbled upon... @Pierre what the answer means is that during the randomForest procedure the train data is resampled with replacement over and over again (AKA bootstrapping), thus if many bootstrap samples (=many trees) are learned, you can switch to OOB once the error converges. $\endgroup$ – Adi Sarid Sep 4 at 5:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.